GSMS-B/Indian-Legal-Qwen2.5-3B

TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 22, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Indian Legal Qwen 2.5 — 3B by GSMS-B is a 3.1 billion parameter language model, domain-adapted from Qwen 2.5 3B with a 32768 token context length. Fine-tuned using QLoRA on 6,354 structured question-answer pairs, it specializes in India's 2023 criminal justice reform acts: Bharatiya Nyaya Sanhita (BNS), Bharatiya Nagarik Suraksha Sanhita (BNSS), and Bharatiya Sakshya Adhiniyam (BSA). This model excels as a component in RAG pipelines and legal education tools for Indian criminal law.

Loading preview...

Indian Legal Qwen 2.5 — 3B: Specialized for Indian Criminal Law

This model, developed by GSMS-B, is a domain-adapted version of the Qwen 2.5 3B-Instruct model, specifically fine-tuned for Indian criminal law. It focuses on the three landmark 2023 criminal justice reform acts: the Bharatiya Nyaya Sanhita (BNS), Bharatiya Nagarik Suraksha Sanhita (BNSS), and Bharatiya Sakshya Adhiniyam (BSA).

Key Capabilities & Training

  • Domain Adaptation: Fine-tuned using QLoRA on a structured dataset covering all 1,059 sections of BNS, BNSS, and BSA 2023.
  • Comprehensive Coverage: Trained on 6,354 instruction-format question-answer pairs, with 6 distinct question types per section (definitions, scenarios, legal elements, exceptions, consequences).
  • Base Model: Utilizes unsloth/Qwen2.5-3B-Instruct-bnb-4bit as its foundation.
  • Model Variants: Available as a fully merged model for out-of-the-box inference, a LoRA adapter for lightweight loading, and a GGUF quantized version for CPU/Ollama usage.

Recommended Use Cases

This model is designed to function optimally as a component within larger pipelines rather than a standalone answer engine. It excels in:

  • RAG Pipelines: Pairing with a retriever over BNS/BNSS/BSA texts to provide grounded, citation-backed answers.
  • Legal Chatbot Backends: Serving as the generation backbone for legal assistant applications.
  • Legal Education Tools: Building interactive Q&A applications for learning the 2023 criminal justice reforms.
  • Section Lookup Assistants: Combining with a section index to identify relevant legal provisions.
  • Further Fine-tuning: Acting as a starting point for more specialized legal adaptations.

It is not recommended for standalone legal advice or high-stakes legal decisions without qualified human review, nor for jurisdictions outside the specified Indian acts.

Finetunes
3 models